International Journal of Advancements in Research & Technology, Volume 4, Issue 2, February -2015

ISSN 2278-7763

40

A New Approach to Determine the Classification of

Mammographic Image Using K-Means Clustering Algorithm
S.Julian Savari Antony1 ,Dr.S.Ravi2
1

Research Scholar, Department of Electronics and Communication Engineering, Shri J.J.T. University, Jhunjhunu, Rajasthan, India, 2Faculty of
Electrical Engineering, Selvam College of Technology,Namakkal, Tamil Nadu, India
Email:savari.sm@gmail.com

ABSTRACT
Breast cancer is one of the most emergent disease in women. Image classification is a supporting for medical system as well as
the difficult task also. Mammographic images contain various features like space, distance, circumscribed masses and microcalcification. This paper addresses on classifying mammographic images created on the Region of Interest (ROI). The images obtained may be blurred due to the variation in illumination, intensity and contrast which cannot be directly processed. It will be
discussed two stages for classify the image. At first stage, the quality of image is increased using histogram equalization method
which normalizes the image. At second stage, the intensity features are computed from the image of shape features, region features such as 'Solidity', 'Eccentricity', 'Convex Area', 'Orientation', 'Perimeter', 'Major Axis Length', 'Minor Axis Length' which
are extracted to compute volumetric values. The proposed classification method is reduced by the false positive result arise in
mammogram classification and overcome missing features while using single feature mammogram classification. Therefore, the
quality of the image classification is a final aim of this paper that is improved. The result shows that the proposed method has
increased the classification accuracy.

IJOART

Keywords : Mammogram image; Region of interest;Feature extraction;K means clusterin;:Time complexity

1 INTRODUCTION

reast cancer is one of the major causes of death among

women. Small clusters of microcalcifications looking as
collection of white spots on mammograms show an early
warning of breast cancer. Primary deterrence seems impossible since the causes of this disease is still remain unknown. An
improvement of early diagnostic techniques is critical for
womens quality of life. Mammography is the main test used
for screening and early diagnosis. Early detection performed
on X-ray mammography is the key to improve breast cancer
prognosis. In order to increase radiologists diagnostic performance, several computer-aided diagnosis (CAD) schemes
have been developed to improve the detection of primary signatures of this disease: masses and micro calcifications. Masses are space-occupying lesions, described by their shapes,
margins, and denseness properties. A benign neoplasm is
smoothly marginated, whereas a malignancy is characterized
by an indistinct border that becomes more speculated with
time. Because of the slight differences in X-ray attenuation
between masses and benign glandular tissue, they appear with
low contrast and often very blurred. Micro calcifications are
tiny deposits of calcium that appear as small bright spots in
the mammogram.Clustering is a division of data into groups
of related objects. Every group, called a cluster, consists of
objects that are similar between themselves and different
compared to objects of other groups. Cluster analysis is a main
task of Data Mining. It divides the datasets into several meaningful clusters to reflect the data sets' natural structure. Cluster
Copyright 2015 SciResPub.

is aggregation of data objects with common characteristics

based on the measurement of some kind of data. There are
several commonly used k-means algorithm, hierarchical clustering algorithm, self-organizing maps algorithm, expectation
maximization clustering algorithm and Fuzzy C means. All
these algorithms are compared according to the following factors: size of dataset, number of clusters, type of dataset, intensity based segmentation and type of software used. Some
conclusions that are extracted belong to the performance,
quality, and accuracy of the clustering algorithms [1],[2].In [3]
aims to develop a software that can define the stage of breast
cancer based on the size of the cancerous tissue. Steps of the
research consist of mammogram image acquisition, determining the Region of Interest, using Region growing segmentation
method [12], measuring the area of suspected cancer, and determine the stage classification of the area on the mammogram
image by using Sample K-Means Clustering method.
In this paper two types of classifications support vector
machine and linear discriminant analysis are used to analyse
the mammographic images. The two classification methods
are using the image pre-processing in wavelet decomposition
and image enhancement [4].There are several stages to breast
image processing. The first stage, breast image acquisition
through mammography. The next stages are pre-processing
image, segmentation, feature extraction, feature selection and
classification [5]. With technique digital mammography, charIJOART

International Journal of Advancements in Research & Technology, Volume 4, Issue 2, February -2015
ISSN 2278-7763

acteristics calcification, circumscribed, speculated and other

ill-defined masses can be diagnosed [6].The paper [7] has following work plans: at first the quality of image is increased
using histogram equalization method. At second, the intensity
features are computed from the image. The classification of
mammograms with benign, malignant and normal tissues
using independent component has been studied in [8].Optimal
set of features selected by Genetic algorithm are fed as input to
Adaptive Neuro fuzzy inference system for classification of
images into normal, suspect and abnormal categories [9].
Detection of edges in an image is a very important step towards understanding image features. Since edges often occur
at image locations representing object boundaries, edge detection is extensively used in image segmentation when images
are divided into areas corresponding to different objects. This
can be used specifically for enhancing the tumor or cancer
area in mammographic images. Different methods are available for edge detection like Roberts, Sobel, Prewitt, Kirsch and
Laplacian of Gaussian edge operators [10], [13].
Region
Growing is a procedure that classifies the pixels or sub-regions
into larger regions based on predefined criteria. The approach
basically starts from the beginning of the set of points, then the
area is enlarged by adding each neighboring pixel point that
has properties similar to those points (for example the range of
intensity or color specification). The selection of similar criteria, in addition to depending on the problem at hand, also
depends on the type of image data available, for examples of
descriptors include moment and
texture [11].
The aim of this paper is to propose a new approach to
determine the classification of Mammographic Image Using
K-Means Clustering Algorithm. This work is organized as follows. In Section 2, it present the proposed method includes the
process of classification. Next, in Section 3, the results are
shown. Finally, Section 4 presents some concluding observations.

41

Input
Image

Euclidean
Distance
Preprocessing

Volumetric
For
Mammographic
Image

Feature
Extraction

Label Assignment
Mammographic

Volumetric
computation

Result

IJOART

2 PROPOSED METHOD
The proposed method has following work plans in figure. 1
system Architecture. The quality of image is increased using
histogram equalization method which normalizes the image.
Then the intensity features are computed from the image and
then shape features, region features like 'Solidity', 'Eccentricity', 'Convex Area', 'Orientation', 'Perimeter', 'Major Axis
Length', 'Minor Axis Length' are extracted to compute volumetric values. Computed feature set is used to classify the image using k-means clustering.

Figure.1. System Architecture

2.1 A. Pre-Processing Final Stage

The process of removing noise present in the image will be
removed in this stage. It shows in figure.2 input image which
is applied with the Gabor filter to remove the noisy signals
present in the image. The noise removed image is applied
with the histogram equalization techniques in figure.3 which
normalize the intensity values of the image. The pre-processed
image has uniform features with higher intensity value persists. A 2-D Gabor function is a Gaussian modulated by a sinusoid. It is a non-orthogonal wavelet. Gabor filters exhibits
the properties as the elementary functions are suitable for
modelling simple cells in visual cortex and gives optimal joint
resolution in both space and frequency, suggesting simultaneously analysis in both domains. The definition of complex Gabor filter is defined as the product of a Gaussian kernel with a
complex sinusoid.

Figure.2. Original image

Copyright 2015 SciResPub.

IJOART

International Journal of Advancements in Research & Technology, Volume 4, Issue 2, February -2015
ISSN 2278-7763

42

tween feature vectors the classifier assigns a label for the submitted test image.

3 RESULTS AND DISCUSSION

Figure.3.Histogram equalization image

2.3 Feature Extraction
The feature extraction has provide to accurate recognition and
feature patterns. In particular, the significant features must be
encoded in this experimentation and used to extract the intensity features which is the Gabor filter with low, high frequencies and four different orientations. Three low frequencies and
three high frequencies with four orientations give 12 combinations of Gabor filter. In order to extract the region properties
have implemented by Gaussian filter. Then the filtered image
is converted into binary image that is identified by the regions
and the features named solidity, eccentricity, convex area, orientation, perimeter, major axis length, and minor axis length
are extracted. The extracted features are stored in a feature
vector which will be used in further analysis.

The proposed methodology has been evaluated with various

data sets of mammographic images. The proposed method
uses different data set and various numbers of classes namely
classification, circumscribed masses, speculated masses, illdefined masses, architectural distortion and asymmetry.
Number of samples has been used for each class. The following Table I shows the data set used for the evaluation of the
proposed method. The Department of Defense (DoD) data
base for breast cancer research program of Standford School of
medicine, California and Mammographic Image Analysis Society (MIAS) data set and Digital Data set for Screening
Mammography (DDSM) for the evaluation of the proposed
method. The proposed method uses less number of samples,
i.e. less than 10% of training samples. Also it takes less time for
training and testing, because unlike other methods, only few
stages are used to identify the mammogram and image classification

IJOART

2.4 Volumetric Estimation

From computed region and shape features the volume of the
shape is calculated using multiple points of the region. Using
the distance metrics calculate the overall volume of the component which is calculated the different co-ordinate points.
The boundary points are identified and split the points into
simple shapes less than polygon. Based on the separated multi-dimensional space, for each shape volume will be compute.
2.5 Mammographic Classifier
It classifies the mammographic image and generated feature vectors which are evaluated by each feature such as intensity value, density measure, solidity, eccentricity, convex
area, orientation, perimeter, major axis length, and minor axis
length. The feature vector has estimated to number of intensity
values, density measure and according to the feature vector is
the size of the pixels covered under the shape or region. Its
mean value of intensity and density is computed based on the
pixel values present in the region. Rests of the features are left
as it is to calculate the Euclidean distance by the k-means clustering.
The mammographic classifier computes the Euclidean
distance iteratively to evaluate the relative measure between
the feature vectors present in the training and testing set.
Based on the mean values computed with different class beCopyright 2015 SciResPub.

TABLE 1
SHOWS THE DETAILS OF DATASET USED

Database

DoD BCRP
MIAS
DDSM

Number of
samples
750
322
2640

Number of testing images

180
115
43

It is shown in Figure.4. Process Flow of K-Means Clustering

Algorithm which is focused into the feature vectors. The database is applied to the K-means clustering which is produced in
Figure.5 shows the identified interest points and it is clear that
the region marked with red colour represents the presence of
cancer cells.

Intensity value
Density measure
Solidity
Eccentricity
Convex area
Orientation

K -Means Clustering
Algorithm

Perimeter
Major axis length
Minor axis length
IJOART

International Journal of Advancements in Research & Technology, Volume 4, Issue 2, February -2015
ISSN 2278-7763

43

Figure.4. Process Flow of K-Means Clustering Algorithm

Figure.5. Identified interest points

The Table 2. shows the percentage of classification accuracy
against five different algorithms. It clearly depicts that the
higher classification accuracy of 99% is attained through the
proposed K means clustering has produced higher classification accuracy than the existing methods taken for comparison.
It is clear that the proposed method produces higher accuracy
at all type of class combinations.
TABLE 2
EVALUATION OF CLASSIFICATION ACCURACY
Algorithm

IJOART

K-Means
SVM
Neuro Fuzzy
Fuzzy C Means
K Means Clustering

Classification
Accuracy
95
96
97.2
98.1
99

Graph2: shows the time complexity of different methods

The graph2 shows the time complexity produced by different
algorithms. It is clear that the proposed multi-variant approach has produced less time complexity which shows the
more classification accuracy.

Graph1: shows the classification accuracy.

The graph1 shows the classification accuracy of different algorithms and it is clear that the proposed multi-variant approach
has produced higher classification accuracy compare to other
methods.

Copyright 2015 SciResPub.

CONCLUSION

This proposed a new approach to determine the classification

of mammographic image using k-means clustering algorithm
which uses different features of the image like intensity values, shape and region features and density features to compute the feature vector. They have computed mean values of
the intensity values of the pixels in the region extracted to
compute the intensity mean value and the density measure is
also computed in the similar fashion. The region metric is
computed using the extracted region values and it has seven
different features hidden. Based on the computed feature vectors the k-means clustering is used to identify the class of the
input image. The proposed system produces good results and
reduces the space and time complexity. It has produced classification accuracy up to 99 % which is more than other methodologies in this era. In future, we will consider other clustering models for feature extraction in order to improve the classification rate.

REFERENCES
[1]

Osama Abu Abbas, Jordan, Comparisons Between Data Clustering Algorithms, The International Arab Journal of Information Technology, vol. 5, no.
3, pp.320-326,Jul. 2008.

IJOART

International Journal of Advancements in Research & Technology, Volume 4, Issue 2, February -2015
ISSN 2278-7763

[2]

[3]

[4]

[5]

[6]

[7]

[8]
[9]

[10]

[11]

[12]

[13]

44

S.Saheb Basha, Dr.K.Satya Prasad, Automatic Detection Of Breast Cancer

Mass in Mammograms using Morphological Operators and fuzzy c means
clustering, Journal of Theoretical and Applied Information Technology,
Vol.7. No.6,2005.
Karmilasari, Suryarini Widodo, Matrissya Hermita, Lussiana ETP, Sample
K-Means Clustering Method for Determining the Stage of Breast Cancer Malignancy Based on Cancer Size on Mammogram Image Basis, (IJACSA) International Journal of Advanced Computer Science and Applications, Vol. 5,
No. 3, 2014
S.Julian Savari Antony and S.Ravi, Development of Efficient Image Quarrying Technique for Mammographic Image Classification to Detect Breast Cancer with Supervised Learning Algorithm, Proceedings of the IEEE Xplore on
ICACCS, pp.1-7, Dec. 2013.
Bozek, J., Mustra, M., Delac, K., and Grgic, M., A Survey of Image Processing
Algorithms in Digital Mammography, Multimedia Signal Processing and
Communications Studies in Computational Intelligence Volume 231, 2009, pp
631-657.
Yasmin, M., Sharif, M., and Mohsin, S., Survey Paper on Diagnosis of Breast
Cancer Using Image Processing Technique, Research Journal of Recent Sciences, Vol. 2(10), 88-98, October 2013.
S.Julian Savari Antony and S.Ravi, Development of Efficient Image Quarrying Technique for Mammographic Image Classification to Detect Breast Cancer with Supervised Learning Algorithm, Proceedings of the IEEE Xplore on
ICACCS, pp.1-7, Dec. 2013.
R.M.Haarlick, Statistical and structural approaches to texture, Proceedings
of the IEEE, vol.67, no.5, pp.786-804, 1979.
S.Julian Savari Antony,Detected Breast Cancer on Mammographic Image
Classification Using Fuzzy C-Means Algorithm International Journal of Innovations in Engineering and Technology (IJIET), Volume 4 Issue 4 December 2014.
Maitra, I.K., Nag S., Bandyopadhyay S.K., A Novel Edge Detection Algorithm for Digital Mammogram, International Journal of Information and
Communication Technology Research, Vol 2 No.2, February 2012.
Kamdi, S., Image Segmentation and Region Growing Algorithm, International Journal of
Computer Technology and Electronics Engineering (IJCTEE), Vol 2 Issue no.1, October 2011.
Priya, D.S., and Sarojini, B., Breast Cancer Detection in Mammogram Images
Using Region- Growing and Contour Based Segmentation Techniques, International Journal of Computer &
Organization Trends, Vol.3 Issue 8,
September 2013.
S. Julian Savari Antony, Improvement of Efficient Volume Estimation
Mammographic Image
Classification Based on wavelet
Analysis,Vol. - III Issue - IV ISSN 2277 - 5730.

IJOART

Copyright 2015 SciResPub.

IJOART